This is an ops protocol for running multiple AI coding CLIs in parallel in tmux.
Core idea: YAML inbox files as the source of truth (auditable via git diffs) + inotify-driven wakeups (no polling). tmux only gets a
tiny “inboxN” nudge.
I’d love feedback on:
1) mailbox protocol design (what would you change?)
2) failure recovery / redo protocol
3) safe wakeup patterns for TUIs (avoid clobbering human input)
Author here.
This is an ops protocol for running multiple AI coding CLIs in parallel in tmux.
Core idea: YAML inbox files as the source of truth (auditable via git diffs) + inotify-driven wakeups (no polling). tmux only gets a tiny “inboxN” nudge.
I’d love feedback on: